MiniMax Plans New AI Model with 2.7 Trillion Parameters
Chinese AI firm MiniMax (Xi Yu Keji) is reportedly planning to launch a new large-scale AI model with 2.7 trillion parameters. The model, internally codenamed M3 Pro, is expected to enhance complex reasoning capabilities.
Chinese AI company MiniMax, also known as Xi Yu Keji, is reportedly preparing to release a new generation large-scale AI model with an ambitious 2.7 trillion parameters. This move signals a significant push in the development of advanced artificial intelligence.
The new model, internally codenamed M3 Pro, is anticipated for release possibly in the third quarter of this year. According to sources, it is designed to substantially improve the handling of complex reasoning and multi-step tasks. Furthermore, the company plans to make this model open-source, potentially fostering broader AI development.
Founded in early 2022, MiniMax develops its own multimodal models and AI-native applications. The company states its services have reached over 300 million individual users and more than one million enterprise clients and developers across more than 100 countries.
In June 2024, MiniMax released its previous flagship model, M3, as open-source. This model features a total of 428 billion parameters with 23 billion active parameters. MiniMax claimed M3 achieved industry-leading performance in coding and agent benchmarks, capable of autonomous task decomposition, tool utilization, and multi-step reasoning.